An Auto-Calibrated Sense Amplifier with Offset Prediction Approach for Energy-Efficient SRAM

摘要

In this paper, for the first time, a novel offset suppression technique is proposed to tackle the offset issue. The key idea is to improve bit error rate (BER) with an energy-efficient offset prediction-based sense amplifier (OPB-SA) for static random access memory (SRAM). The OPB-SA effectively compensates for the branch current mismatch due to threshold voltage (V-TH) offset in SA sensing devices. Extensive simulation results, referring to an industrial hardware-calibrated UMC 65-nm CMOS technology, show that OPB-SA achieves 27.2, 20 and 11.1% offset reduction over current latch SA (CLSA), SA with inherent offset cancellation (SAOC) and offset-compensated current SA (OCCSA), respectively, without sacrificing performance. The OPB-SA features significant offset suppression capabilities with 31.3, 12.2 and 7% tighter offset distribution compared to CLSA, SAOC and OCCSA, respectively. The energy efficiency is 0.26fJ/bit, thus improving 61.04, 84.16 and 87.12% over SAOC, OCCSA and body bias SA (BBSA), respectively. The OPB-SA requires 0.72x, 0.8x and 0.88x less bit-line swings than CLSA, SAOC and OCCSA for targeted 0% BER. Hence, overall SRAM macro with proposed scheme exhibits a superior dynamic power metric over the conventional designs with 0.66x, 0.74x, 0.98x and 0.81x lower bit-line power consumption than CLSA, SAOC, OCCSA and BBSA, respectively.